To:       Subscribers to CLASS-L
From:   Stan Sclove, Secretary/Treasurer, CSNA

Are you aware of any studies applying clustering techniques to political
redistricting problems?
A colleague is interested in constraints on gerrymandering.

    I'm not sure simple solutions like requirements of convexity, or
restrictions on some measure of eccentricity would even be generally
applicable.
    What would be needed would be a clustering of an arbitrary spatial
distribution into a given number of regions of equal membership, within a
given tolerance.
    The solution need not be unique. The politicians could fight over which
of the acceptable distributions would give them the best political
advantage. Unfortunately now the majority party is essentially
unconstrained.
    Surely this problem must have been addressed. If not, maybe it could
easily be with existing methods.

     The colleague has looked into the parameters of a typical problem:
California has 5858 census tracts,
and the number of election districts runs from 40 (State
Senate) to 80 (State Assembly).
In the case of Congressional districts (60-ish in Calif.), the Supreme Court
has given a population tolerance of 1%.

    Thank you for any suggestions you may have.

-- stan